RF-CH-6.4 SUDHA.pdf
Media
- extracted text
-
’I
A
RF-CH-6.4 SUDHA
THE DETERMINANTS OF INFANT MORTALITY
IN REGIONAL INDIA
By
Michael Beenstock
and
Patricia Sturdy
City University Business School
October 1986
I
Introduct ion
in which
There is a well established empirical tradition
investigate
multiple regression techniques are used to
the socio-economic determinants of life expectancy or
Krishnan
disaggregated mortality rates,
e . g . Da Vanzo (1985),
(1975) and Beenstock (1980).
These studies have typically
been conducted using interna tlona1 cross section data
although there are exceptions where regional aggregates
have been used for a given country.
In the latter case,
regional variations i n mortality rates and various socio
economic variables are studied to determine the factors
that systematically account for these variations.
A
recent
addition to the literature is Jain (1985), who
has investigated the data generated by the Survey of Child
and Infant Mortality which was carried out in India in
1 978.
He regressed infant mortality rates on a range
of socio-economic variables using state-wise data for
rural India.
Ruzicka (1984) has used more informal techniques
for analysing these data.
Here ,
we present our own analysis
of this survey which supplements Jain’s efforts in several
respects.
First, we show that a richer statistical model of infant
mortality can be estimated if the class of causative
variables
is
first converted into factor scores using
factor analysis, see Lawley and Maxwell (1971).
This
approach is useful because many of the causative variables
which we investigate are colinear.
We therefore find
be incorporated
that a broad range of causative variables can
is much
into the model whereas Jain finds that the range
narrower .
2
Secondly ,
educa tion on
t o isolate the effects of e.g .
infant mortal] t y , we propose a more direc t
one proposed by Jain.
test
than the
This consists o f invest, igat ing
separately the infant mortality rate among children whose
mothers are educated and t he infant mortality
children whose mothers are not educated.
rates among
Thirdly,
because
infant mortality rates are naturally bounded between zero
and a thousand,
linear regress ion i s not strictly appropriate
We therefore experiment
with various non- 1 inea r
transformations
of the data on which basis we find that a semi-logistical
model provides a superior description of the data.
well known e . g . (Wy on and Gordon 1971),
states
female
It
is
that. i n certain Indian
infant, mortality rates are greater than that.
of their male counterparts.
One
possible explanation for
this i s that. females respond to the causative variables
(e.g.
education of the mother)in a quantatively different
the female model coefficients happen t o
way to males,
i.e.
be different
from the male model coefficients.
An alternative
hypothesis i s that there i s a genuine preference for male
infants and that the causative variables are not respons ib le
for differences in male and female infant mortality rates.
In Section 11 we propose a methodology which enables us to
discriminate between these competing hypotheses.
These and
related empirical results are reported in Section III.
Methodological issues are described i n Section 11 .
3
II Methodology
Basic Hypotheses
As in e.g. Jain (1985), a stochastic model is proposed in which
the infant mortality rate, at a given point in time, in state i
is hypothesised to be determined by a vector of socio-economic
variables X, i.e.
(D
IMR.i = F(X.,
i ui)
where u^ is a stochastic term, IMR denotes the infant mortality
rate and X i
(^i ’ X2i’
The constitution of X.. is in the last analysis an empirical
matter.
However, following earlier research we experiment with
the following variables that are included in the Survey of Child
and Infant Mortality or are obtained from other sources as
described in the appendix.
X1
availability of medical facilities, % village
with medical facilities greater than 5km
distant (+)
X2
medical attention at birth, % births attended
by trained medical staff (-)
X3
nutrition,
%% population consuming less than
2,100 calories per diem per capita (+)
X4
clean drinking water.
% population using tap
as main source of drinking water (-)
%
4
X5
poverty,
% sample households with per capita
monthly household expenditure below 50 Rupees
(+)
X6
literacy, % of adult female literates (-)
X7
vaccination.
%
female
i nfants
gi ven
DPT
vaccinations (-)
X8
Hindu, % population Hindu (?)
X9
Muslim, % population Muslim (?)
X10
caste, % population in scheduled caste (?)
X11
tribe, % population in scheduled tribe (?)
X12
overcrowding, % households with one room only
(+)
The signs of partial derivations (F.) have been indicated in
parentheses e.g. the larger the proportion of the population not
covered by medical facilities, the higher is likely to be the
infant mortality rate.
In the past, it has proved difficult to
estimate equation (1) because of high degrees of colinearity
between many of the causative variables.
our own data.
Similar problems beset
Therefore, instead of estimating equation (1) we
estimate
IMR = GUj-j, vi)
(2)
5
where v .
i
is a random disturbance term and
K
X, .
w
jk ki
Z .
Ji
i s the factor score of the J ' th factor and w .. are the
Jk
(rotated) factor loadings, see e.g. Maddala (1977)
It
follows from this that
(SIMP
6X.
k
_6F = Z.L w . 6 0
J jk6Z .
6X,
k
J
(3)
i s the response of the infant mortality rate to changes
i n the k’th variable.
Factor analysis of the K variates generates J significant
factors (where J
K ) which are orthogonal to each other.
Thus equation (2) consists of only J independent regressors
rather than K correlated regressors
as i n eguation (1).
Below we report various estimates o f egua t ion ( 2 ) and
calculate the partial derivatives in terms of eguation
(3) .
Functional Forms
Previous research has on the whole paid little attention
to the functional form of equation (1).
equation (1)
i n terms of a linear model,
m ortality rate i s naturally
1 ,000.
Before these natural
Jain e.g.
estimates
yet the infant
bounded between zero and
limits are reached it seems
appropriate that the estimated model should rule out very
high or very low infant mortality rates.
on fig.
This i s illustrated
1 where the crosses represent e.g. cross section
observations of state-wise infant mortality rates with
respect to some positive valued variable,
X.
Linear regression
6
IMR
1 000
a
0
Choice of Functional Form
F ig.
would generate a regression line such as b) which implied
that the infant mortality rate could either be greater
than 1000 or below zero.
In contrast,
schedule a ) implies
that the in fant mortality rate has unknown upper and lower
limits that are below 1000 and greater than zero respectively.
Moreover ,
form,
if schedule a) is indeed the appropriate functional
the linear model will generate inefficient parameter
estimates, because it will not account for the outlying
observations.
Below we hypothesise a logistical function of the type:1 n IMR/1000
1-IMR/100
a
+ u
(4)
7
which implies that the infant mortality rate is asymptotical ly
bounded by 1000 (1 _(_e
)-1
and 1 000 .
Indeed,
we find this
fits the data better than the linear model.
Data Control
Suppose we wish to estimate the effect o f educa tion on
infant mortality using the adult literacy rate as an appropriate
One way of doing this is to estimate eguation
proxy .
(1) where one of the X variables is the literacy rate
and where IMR is defined for the population a s a whole.
A second way is to define IMR i n terms of the literacy
of the parents,
to omit the literacy rate as a regressor
and then to estimate the following models
/ IMRh/1000
. i Xi
1n
ah + 3 hk
k + uh
I 1-IMRh/l000
h-1 ,
(5)
2
where
IMR1
infant mortality rate of illiterate parents
imr2
infant mortality rate of literate parents
If indeed literacy lowers the infant mortality rate, we
should find that a2
Fortunately,
0t1 .
the Survey of Child and Infant Mortality
controls the data in this way.
Indeed,
it controls 11
for
the literacy of the mother as well as a range of other variables
including source of drinking water,
employment, status of mothers.
age at marriage and the
However these controls are not
integrated, 1.e. we cannot distinguish literacy and drinking
water supply simultaneously.
We may therefore compare and
contrast statistical models of infant mortality for different
sub-groups
of the population.
In this way we are likely to obtain
a clearer picture of whether literacy etc. exerts an independent
effect on infant mortality.
8
female
the
of whether
the male one .
female
the
tests of alternative hypotheses
lends itself to
This methodology
In certain
infant
mortality
infant
Indian States,
mortality
rate appears
This may e 11 her be due t o a pro-male
the
purposes
the
as
for
represented by equations
IMRf
IMRm
IMRf
female
IMRm
male
Equations
in
the
the
an-
may
it
implicit.
reflect
in
the
for
illustrative
variables X,
and
X2
T
and
and
for males and
fema les
(7)
+ 01 Q
(6)
6
* 62X2
+ BD
(7)
X
1 ,X1
(a 1
IMRm
rate s a s
follows:
- Bj^l
indicates
insofar
rate.
imply that we can de compose differences
(7)
t hat
-b 2 )X2
i n any
as ot'i exceeds gq
three possible
and
Ct2
exceeds $2
the
rate
whatever
the values of the parameters
for
rate may exceed the male
variables happen
the
infant mortality
t o assume particular
there
i s an all
India
cr = B
2 2
the
infant mortality
female
female
infant
given values of the causative variables.
mortality
pro-boy bias,
in
factors.
rate will be greater than the male
infant mortality
(8)
Bo
Cl o
particular state differences
rates reflect
infant mortality
first
high.
relatively
+ “2X2
infant mortality
(8)
i n our d a t. a
“l X1
infant mortality
Equation
indeed
than
infant mortality rate
(6)
IMRf -
(6)
i s greater
Suppose
mortality models were different
infant
India
to be
are
infant mort alit y.
there were two causative
and
bias o r
pa ramet ers and the variables that
causat ive model
in
rate
Secondly
female infant mortality
rate
values.
since even
if the X
Thirdly
if ao>6o
if a1 = B>| and
rate would exceed its male
counterpart.
Below we
ot0
is
implement
this methodology and test the hypothesis
significantly greater
estimates of
Cl
and
than Bo
.
We also estimate
that
independent
9
111
Empirical
Results
Jain's Model
Jain (1985) v iewed the problem of infant mortal ity at
the household level
three levels: the village level ,
and the individual
level.
Using the same variables as
Jain, we were essentially able to replicate his results.
shows how infant mortality may b e explained by
Table 1
DPT vaccination, poverty and female literacy .
However,
it was possible to improve the fit of the model by the
inclusion of caste (model 2).
that Jain
found
It
is
interesting to note
a positive relationship between the presence
o f medical facilities and the infant mortality rate yet
when usage of medical facilities was considered it showed
a st rong negative correlation.
Therefore we regard our
model 2 in table 1 as a substantial advance on Jain's
efforts.
Nevertheless,
these results were somewhat disappointing,
as we were unable to estimate the possibl
by other independent variables.
part played
For example,
neither
Jain nor ourselves were able to bring the 'clean drinking
water'
variable into the model.
Initial bivariate regressions
gave a Pearson coefficient of 0.061.
However, through
the factor score approach it was possible to estimate
in fant
the importance of tap water and other variables upon
mortality.
ii
10
Table 1
on it
Jain Original Multiple Regress ion Model with Other Models Based
IMR 1978
Dependent variable
Independent
variables
Jain’s
Model 8
High school
-0.079
DPT vaccmat ion
Model 1
Model 2
a
_B__
a
-0.576*
-2.181
(1.817) -1.992
(0.959)
Poverty
0.437*
0.87
(0.432 )
0.924
(0.311)
Adult female
literacy
-0.443*
-0.073
(0.656) -0.56
(0.362)
Birth attendance
-0.008
0.089
(0.765)
Presence of
medical facilities
0.176
0.245
(0.398)
% sick children
seen in medical
institutions
-0.331
(0.285)
Caste
1 . 641
(0.516)
0.682
0.847
R2
0.524
0.77
Standard error of
estimate
F
21.83
15 .152
4.3
11.09
109.94
103.46
R2
0.765
4.34
Constant
* Statistically significant at p - 0.1
B
regression coefficients
a
standard error of B
11
’ i ~z
.5%^
The Factor
A
Score Models
factor model was estimated
four
explanatory
a s a data
variables
choice
reflected the degrees of
data.
However,
model
i s more suitable.
this analysis.
twelve
This
exercise.
freedom available in the
a more parsimonious
Table 2 shows the results of
Factors
2 and 3 contain a t
1 ,
factor
female literacy and birth
The
reduction
we have not explored whether
variable with a high
water
from eleven o r
loading.
least
For example,
attendance in F actor
1 ,
one
adult
and tap
i n Factor 3 .
factor scores
for each of 1 6 states were calculated,
i n our analysis of infant mortality
and used a s regressors
rates .
The model was o f the logistical
i n section
I.
The
results are shown
in
type described
table 3.
variables respectively ,
A and B contain 12 and
1 1
with the exception o f
'poor nutrition',
Models
and
all the signs
concur with a priori expectations.
Although a better
fit was achieved by the reestimation
of Jain's eguation
through the drastic dropping of variables, here we present
a model containing many independent
we can estimate
variables
from which
the importance of each i n relation
to
in fant mortality.
The multipliers
3,
i .e.
i n table 3 were computed from equation
the estimated parameters of the regression model
were multiplied by the relevant factor loadings.
These
multipliers measure the effect of the underlying variables
on
the logarithm of the relative probability of dying
to living.
For example,
in equation A, if there is a
12
1% increase in children who are vaccinated,
probability of dying falls by 23%.
the relative
These multipliers
therefore have a dimension of an elasticity.
They estimate
the percentage response o f the relative probability of
dying to the percentage cover of the explanatory variables.
The factor score models confirm the findings of Jain and
ourselves insofar as they show that DPT vaccination,
literacy and poverty are important
in infant mortality.
Model A shows how the lack of medical
births play an important part.
female
facilities and attended
For instance,
for every
1% increase in the births attended by trained medical
staff,
the relative probability of dying falls by 18%.
The factor score approach enabled us to show that clean
drinking water could influence infant mortality .
For every 1% increase in the population drinking.tap water
the relative probability of infants dying falls by 10%.
Religious and sociological factors also appear to be important
determinants.
However,
it would be difficult to speculate
upon the reasons behind these findings.
The relationships
between caste and infant mortality have been investigated
within the control groups.
There was a positive relationship
between people living in crowded housing conditions and
infant mortality.
It may be that in the rural situation
crowding is less important than in the urban environment.
For every 1% of households,
living in one room,
the relative
probability of infants dying falls by only 2 to 6%.
It is noteworthy that the poor nutrition variable was
the only one to have an unexpected sign.
The nutrition
13
data were collected i n 1974 by the National Sample Survey
and were included to give a more complete picture of the
problem.
However,
it may be that there was accurate reporting
i n the educated states, such as Kerala, with the result
'poor nutrition'
that
would go hand i n hand with high
female literacy rates.
Factor analysis shows that poor
nutrition i s always grouped with adult fema 1e literacy
(table 2 ) .
Factor Score Models
Table 2
Factor Analysis of Independent Variables
Factor Loadings (w
)
Factor 2
Factor 3
Factor 4
a 11 en da nee 0.90 1 8 3
-0.0627
0.21902
-0.22269
Lack of medical-0.14666
facilities
Vaccination
0.5482
0.7805
-0.5066
0.0410
-0.1124
0.7604
-0.0241
Variable
Birth
Factor 1
Adult female
literacy
Poverty
0.8441
-0.36613
0.1448
0.0327
0.150
0.3267
-0.618
0.319
Tap water
0.0205
0.2385
0.8341
0.0650
Hindu
-0.238
0.7884
-0.115
0.3088
Muslim
0.1093
-0.6131
-0.0414
0.2829
Caste
-0.0200
-0.009
0.0313
-0.8976
Tribe
-0.308
0.260
-0.430
0.3985
Poor nutrition
0.8102
0.34118
-0.2322
0.2793
Crowding
0.1781
0.6833
0.1273
0.1378
FACTOR
1
2
3
4
Cumulative percentage
variance explained
48.3
76.4
89.4
100.0
chi-square
171.71552
120.59752
92.66852
53.40533
14
Table 3
Logistical Regressions Using Factor Scores
Dependent v ariable In( IMR/1000-IMR)
Model A
B
a
Model B
a
-2.004
Constant
-2.010
S1
-0.164
(0.057)
-0.167
(0.0533)
S2
0.124
(0.057)
-0.189
(0.063)
S3
-0.168
(0.067)
0.130
(0.054)
S4
-0.056
(0.061 )
-0.409
(0.057)
R2
0.5-5
0.61
a
0.218
0.20
Constant
-2.011
-2.006
S1
-0.168
(0.0571)
-0.170
(0.052)
S2
0.122
(0.057)
-0.185
(0.062)
S3
-0.164
(0.066)
0.127
(0.0532)
R2
0.56
0.62
Standard
error of
est imate
0.217
0.20
Variables
Multipliers
-18
Birth attendance
-18
Lack of medical facilities
20
DPT vaccination
-23
-23
Adult female literacy
-21
-22
10
10
-11
-13
-Hindu
13
12
-Muslim
-10
-10
•Caste
4
4
Tribe
13
-6*
12
-5*
Poverty
Tap water
Poor nutrition
Crowding
* Denotes unexpected sign
2
6
-15-
Controlled Data
Table 4 gives the results of factor score models for control
groups .
To test for the effect of literacy we now implement
equation (5) to see whether the constant term is signifi
cantly higher for illiterate mothers than for literate mothers,
reported in table 4 equations A and B.
We find that the
constant term for literate mothers is indeed smaller and
suggests that the relative probability of infants dying in
the case of literate women is 44% smaller than for illiterate
women.
the model fits better for literate than illiterate
mothers .
Adding the general female literacy as a regressor
made little difference to either model,
thus implying that
the general literacy of females at the community level does
not apear to affect the infant mortality rates of either
literate or illiterate mothers.
A comparison was made between mothers who drink tap water
and mothers who use other sources of drinking water (Table
4 equations E and F.
Again, the constant term in the tap
water model was lower indicating that the relative probability
of infants dying for mothers drinking "unclean water" was
11 % higher than for mothers who drink tap water.
However ,
this difference was not statistically significant.
The infant mortality rate in mothers whose age at marriage
was under eighteen was directly compared with the infant
mortality rate of mothers whose age at marriage was over
twenty one.
Table 4(equations I and Jjshows how when controlling
for other independent variables the relative probability
Fable 4:
Factor Score Models for Control Groups
16
Dependent Variable: Ln((IMR Control)/1000-IMR(Contro
Variable
Birth attendance
Lack of Medical Fac
M
0 Vaccination
L
T Adult Female Literacy
I
p Poverty
L
I Tap
E
R Hindu
S
Muslim
Relative Probability of Dying/Living
A
IMR in
Literates
B
IMR in
Illiterates
C
IMR
Caste
D
IMR
Hindu
E
IMR
Tap
F
IMR
Non Tap
G
IMR
Workers
H
IMR
Non-Wo
-20
-16
-15
-21
1.6*
-22
-14
-25
31
17.5
17
20
7
21
13
21
-42
-19.2
-16
-25
-12
-25
-18
-27
-14
-25
4.5*
-27
-15
-29
12
6
7
6
-10
-11
31
4.8
6.3
9
-36
-7
-7
-12
17
12.5
9
-2
-12.3
-3.5
Caste
8
9.8
Tr-.be
28
9.7
Poor Nutrition
0.7
Crowd]ng
-3
14
8
14
-11
5.2
-13
-5
-12
7
3.5
10
2
8
11
13
4.6
12
10
14
-10*
7
-1*
6
-14*
-6*
16
-1*
0.82
-1.6*
2
-8*
0.5
-0.7*
0.5
0.59
0.51
0.067
0.62
0.31
0.53
0.08
0.54
Standard error estimate 0.33
0.23
0.316
0.22
0.22
0.28
0.32
0.29
Constant
-2.437
-1.993
-1.943
-2.058
-2.175
-2.054
-1.929
-2.207
0.0894
0.0622
■ 0.0666
I
0.0749
,0.864
0.077
R2
o
Sign]ficance +
2r7
Oit
+ Ng is significantly different at N. Standard deviations
3 Coefficients of 4 Factor Score regressors are
/ailable on reguest.
1cr
.)
-17of infants dying is 38% greater when mothers marry before
the age of eighteen.
The effect of female participation upon infant mortality
was investigated with controlled data.
Again the constant
term was significantly higher for working mothers,
thus ind-
icating that female employment status exerts some independent
effect upon infant mortality.
The relative probability of
infants dying was 27% higher among working mothers
(Table
4 eguations G and H).
The factor approach failed to explain the infant mortality
rate of scheduled caste mothers (eguation C).
The reason
for this may be that our factor score model did not contain
variables that were relevant to infant mortality in this
group.
Similarly,
for
'mothers drinking tap water',
the
model was also weak; in this particular model some of the
signs on the relative probabilities were not as we would
have expected.
However,
for mothers drinking non-tap water,
the factor score model was stronger.
In fact,
this model
and the Hindu controlled model could be readily explained
by the factor score approach and were not dissimilar to model
A.
Table 4 (eguations K and H) also compares the male and female
infant mortality models.
The constant terms were not sig-
nificantly different thus indicating that there was no apparent
bias in favour of male infants at the all India level.
-18-
Functional Forms
The goodness of fit of logistical and linear models cannot
be directly compared since the dimensions of the residuals
are quite different from each other.
The equation standard
error of est imate from the linear four-factor model was 21.95
expressed in unit rates of infant mortality,
Table 5
whereas the
Comparison of Linear and Logistical Models
F
R
estimated transformed
standard
standard
error
error
Linear
4.73
0.516
21 .95
21 . 95
Logistical
5.42
0.55
0.219
21 . 3
(IMR)
standard error in the case of the logistical transform of the
infant mortality rate was 0.219.
To see whether the logistical
model is a better description of the data, we have appropriately
transformed the fitted values of model into units of infant
mortality and calculated the adjusted standard error of
the transformed residuals.
19
As indicated i n table 5,
this turned out to be 21.3, which
i s smaIler than its linear counterpart of 21.95.
an F ratio test indicates that a t P
However,
0.05 these two
standard errors are not statistically significantly
from one another.
different
On the other hand it i s
noteworthy that the logistical model,
suggest,
fits the data better,
as
intuition would
but substantially larger
samples would be necessary t o establish this at conventional
levels of confidence.
Conclusions
1 .
A re-estimation of Jain's regression model showed that
DPT vaccination, adult fema 1e literacy, poverty,
caste and
the usage of medica1 facilities were important determinants
of infant mortality i n rural India.
A better fitting model
was achieved by dropping a number of his independent
variables and adding others.
2.
To assess the relative probability of dying to living
attributed to each of the independent variables, a fact o r
score model was estimated, in which all the signs were
correct except for
'poor nutrition' .
In addition to those
-20variables indentified by Jain,
it was possible to show
that tap water, birth attendance and sociological factors
were important influences upon infant mortality.
3.
Control groups were investigated by factor score model-
ling technigues.
From these, we conclude firstly, that
the illiteracy of mothers is a contributory factor to
infant mortality and secondly,
that the risk of infants
dying is significantly higher when mothers marry before
the age of eighteen.
Mothers who work were also shown
to run an increased risk of their infants dying.
However
no significant difference could be shown in infant mortality
rates when mothers with different sources of drinking
water were compared.
4.
A comparison between male and female infant mortality
rates was made through the factor score model.
Using
this approach no bias in favour of male infants could
be detected at the all India level.
5.
The choice of functional form was studied.
A logistical
model of the infant mortality rate was shown to explain
the data more accurately than the linear model.
However ,
these differences were not statistically significant.
21
The Data Appendix
I Survey on Infant and Child Mortality,
1 979,
General, Ministry of Home Affairs, New Delhi,
The Registrar
India.
Rural data were collected from 18 of the larger states
of India.
The Survey formed a sub - sample of the Sample
Registration Survey and covered more than 500 thousand house-
holds and included over 3 million people.
The data were
collected on a state-wise basis by a non-medical enumerator.
Vital rates from Bihar and West Bengal are generally regarded
as unreliable and were therefore excluded from the Survey
(Jain,
1985).
The following data were used : -
1 .
his
The Infant Mortality Rate 1978.
dependent variable.
Jain used this for
We also used I MR 1978 to confirm
Jain’s findings and to estimate models 1 and 2 .
There
were ovservations on 16 states.
2.
The Average Infant Mortality Rate.
An average infant
mortality rate was derived from the results of the Sample
Registration Survey between 1972 and 1976, and included
in the 1978 IMR;
This was considered to be more accurate for the
dependent variable.
factor score models.
This infant mortality rate was used for the
- 22 3.
The infant mortality rate among literate
IMR Literate.
mothers.
4.
IMR I lliterate.
The infant mor taiity rate among illi-
terate mothers.
5.
The infant mortality rate among scheduled
IMR Caste.
caste mothers.
6.
The infant mortality rate among Hindu
IMR Hindu.
mothers .
7.
IMR Tap.
The infant mortality rate among mothers
using tap or hand pump as main source of drinking water.
8.
IMR Non-Tap.
The infant mortality rate among mothers
not using tap water as a source of drinking water.
9.
The Presence of Medical Facilities.
medical facilities less than 2km distant.
10.
The Absence of Medical Facilities.
% villages with
(Table 1 only)
?o villages with
medical facilities more than 5km distant.
11 .
Usage of Medical Facilities.
?o distribution of sick
children aged 0-6 years receiving attention in medical
institution.
(Table 1
only) .
23
12.
Tap.
% population using tap as main source of drinking
water .
13.
% sample households with per capita monthly
Poverty.
household expenditure below 50 Rupees.
14.
Adult Female Literacy.
%
literacy i n females over the
?0
literacy i n all females.
age of 15 years.
15.
Total Female Literacy.
16.
Vaccination.
% female infants rece i v mg DPT
vaccination.
% households with one room only.
17.
0ve r-Crowd i ng.
18.
Muslim.
19.
Hindu.
?0
20.
Caste.
% populat ion in scheduled caste.
21 .
Tribe.
% populat ion
22 .
IMR Workers.
mothers.
% population Muslim.
population Hindu.
in
scheduled tribe.
The infant mortality rate
anono working
24 23 .
The infant mortality rate among
IMR Non-Workers.
mothers who do not work.
24.
IMR Age at Marriage before 18 years.
The infant mortality
rate among mothers whose age at marriage was less than
eighteen years.
25.
IMR Age at marriage greater than 21
years.
The infant
mortality rate among mothers whose age at marriage was
greater than twenty one years.
Average Male infant mortality rate 1972,74,76,78.
26.
IMR Male.
27 .
IMR Female.
Average female infant mortality rate
1972,74,76 and 1978.
..-U-IWIT——
- ' -
1
,™"*
25
II
Levels,
Trends and Differentials in Fertility,
The Registrar General, Ministry of Home Affairs,
1979
New Delhi,
India (1982)
Attendance at birth,
this study gave the percentage of rural
b.irths attended by trained medical staff.
18 observations
were made on a state-wise basis.
111
The Sample Registration Survey (1972 to 1976).
The Registrar General, Ministry of Home Affairs,
New Delhi,
India.
This is an ongoing survey which covers 2,400 sample units in
rural areas.
Vital rates were collected from 16 states
(excluding Bihar and West Bengal) on a state-wise basis.
IV
The National Sample Survey
Report number 238.
Round 26 ,
(July 1971 to June 1972).
Volume 1.
1 978.
Calorie and protein values of food items
in rural areas.
The percentage population receiving less
than 2,100 calories per diem per capita,
used as the
' poor nutrition'
observations on 17 states.
variable.
in each state was
There were
26 -
Re ferences
M. ,
Beenstock,
1980, Health, Migration and Development,
Gower Press, Farnborough.
Da Vanzo,
J., Habicht, J-p, Hill, K., Preston,
S. ,
1 985,
Quantitat ive Studies of Mortality Decline in the Developing
World, World Bank Staff Working Paper,
Number 638,
Population and Development Series, Number 8.
Jain, A.K.,
1985, Determinants of Regional Variations in
Infant Mortality in Rural India, Population Studies,
39,
407-424.
Krishnan,
1975, Mortality Decline in India,
Development v .
1951-61:
Public Health Program Hypothesis',
Social
Science and Medicine, 475-9.
Lawley, D.N., and Maxwell, A.E.,
1971 , Factor Analysis as a
Statistical Method, 2nd edition, Butterworths ,
Maddala,
York .
G.S.,
London.
1977, Econometrics, McGraw-Hill Inc., New
i
- 27
Ruz icka,
L.T. ,
1984, Mortality in India: Past Trends and
Future Prospects',
in India's Demography,
Essays on the
Contemporary Population, edited by 1. Dyson and N. Crook,
South asian Publishers Pvt. Ltd., New Delhi.
Wyon ,
J.B. and Gordon J.E.,
1971 ,
Problems in the Rural Punjab.
University Press.
The Khanna Study: Population
Camb r i dge, Mass. Harvard
Position: 2638 (2 views)